To better organize my work, I separated my workflow into different kernels. The submission kernel just imports the trained PyTorch model, and performs the prediction. I thought this would speed up grading also.
\nHowever even that the kernel works perfectly on the example test set, it results in a submission failure. When I use the same code but in one notebook (Training, and prediction), it gets graded correctly.
\nI thought that this is a curious behavior i.e. failure in grading when loading a serialized PyTorch model. Is anybody experiencing this ?
", "messages": [ { "id": "1065633", "postDate": "10/31/2020 14:33:11", "content": "To better organize my work, I separated my workflow into different kernels. The submission kernel just imports the trained PyTorch model, and performs the prediction. I thought this would speed up grading also.
\nHowever even that the kernel works perfectly on the example test set, it results in a submission failure. When I use the same code but in one notebook (Training, and prediction), it gets graded correctly.
\nI thought that this is a curious behavior i.e. failure in grading when loading a serialized PyTorch model. Is anybody experiencing this ?
", "rawMarkdown": "To better organize my work, I separated my workflow into different kernels. The submission kernel just imports the trained PyTorch model, and performs the prediction. I thought this would speed up grading also.\n\nHowever even that the kernel works perfectly on the example test set, it results in a submission failure. When I use the same code but in one notebook (Training, and prediction), it gets graded correctly.\n\nI thought that this is a curious behavior i.e. failure in grading when loading a serialized PyTorch model. Is anybody experiencing this ?", "votes": null }, { "id": "1065644", "postDate": "10/31/2020 14:43:47", "content": "This is strange. Does it give an error immediately or after a few hours?
", "rawMarkdown": "This is strange. Does it give an error immediately or after a few hours?", "votes": null }, { "id": "1065664", "postDate": "10/31/2020 15:11:27", "content": "On average, after 5 - 10 minutes.
", "rawMarkdown": "On average, after 5 - 10 minutes.", "votes": null }, { "id": "1065675", "postDate": "10/31/2020 15:18:58", "content": "Most likely, your notebook doesn't see the \"model file\". When you click Submit , you see the file with weights in the 'Input' section? If you don't see it, the dataset wasn't added …. If you see this file, I don't know :) Then try to save the weights in one Notepad, then read them and make a prediction (to check whether the model is loaded correctly)
", "rawMarkdown": "Most likely, your notebook doesn't see the \"model file\". When you click Submit , you see the file with weights in the 'Input' section? If you don't see it, the dataset wasn't added .... If you see this file, I don't know :) Then try to save the weights in one Notepad, then read them and make a prediction (to check whether the model is loaded correctly)", "votes": null }, { "id": "1065698", "postDate": "10/31/2020 15:49:19", "content": "The file exists in the input folder, I shared with you the submit notebook, and please tell me if you see something that I don't see. I tried to re-sumbit now, with an error again.
\nhttps://www.kaggle.com/abdessalemboukil/submit
\nThanks!
I don't see any reason yet… The only thing I would do is-if your model doesn't use lectures, can only select questions at once? (before 'preprocess(test_data, features)…'
", "rawMarkdown": "I don't see any reason yet... The only thing I would do is-if your model doesn't use lectures, can only select questions at once? (before 'preprocess(test_data, features)...'", "votes": null }, { "id": "1065731", "postDate": "10/31/2020 16:53:43", "content": "It seem not to make a difference in my other notebook. This is an odd behavior.
", "rawMarkdown": "It seem not to make a difference in my other notebook. This is an odd behavior.", "votes": null } ], "comments": [ { "id": 1065644, "author_name": "sapr3s", "author_url": "", "post_date": "10/31/2020 14:43:47", "content": "This is strange. Does it give an error immediately or after a few hours?
", "votes": null, "replies": [ { "id": 1065664, "author_name": "abdessalemboukil", "author_url": "", "post_date": "10/31/2020 15:11:27", "content": "On average, after 5 - 10 minutes.
", "votes": null, "replies": [] }, { "id": 1065675, "author_name": "sapr3s", "author_url": "", "post_date": "10/31/2020 15:18:58", "content": "Most likely, your notebook doesn't see the \"model file\". When you click Submit , you see the file with weights in the 'Input' section? If you don't see it, the dataset wasn't added …. If you see this file, I don't know :) Then try to save the weights in one Notepad, then read them and make a prediction (to check whether the model is loaded correctly)
", "votes": null, "replies": [] }, { "id": 1065698, "author_name": "abdessalemboukil", "author_url": "", "post_date": "10/31/2020 15:49:19", "content": "The file exists in the input folder, I shared with you the submit notebook, and please tell me if you see something that I don't see. I tried to re-sumbit now, with an error again.
\nhttps://www.kaggle.com/abdessalemboukil/submit
\nThanks!
I don't see any reason yet… The only thing I would do is-if your model doesn't use lectures, can only select questions at once? (before 'preprocess(test_data, features)…'
", "votes": null, "replies": [] }, { "id": 1065731, "author_name": "abdessalemboukil", "author_url": "", "post_date": "10/31/2020 16:53:43", "content": "It seem not to make a difference in my other notebook. This is an odd behavior.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1065633": "To better organize my work, I separated my workflow into different kernels. The submission kernel just imports the trained PyTorch model, and performs the prediction. I thought this would speed up grading also.\n\nHowever even that the kernel works perfectly on the example test set, it results in a submission failure. When I use the same code but in one notebook (Training, and prediction), it gets graded correctly.\n\nI thought that this is a curious behavior i.e. failure in grading when loading a serialized PyTorch model. Is anybody experiencing this ?", "1065644": "This is strange. Does it give an error immediately or after a few hours?", "1065664": "On average, after 5 - 10 minutes.", "1065675": "Most likely, your notebook doesn't see the \"model file\". When you click Submit , you see the file with weights in the 'Input' section? If you don't see it, the dataset wasn't added .... If you see this file, I don't know :) Then try to save the weights in one Notepad, then read them and make a prediction (to check whether the model is loaded correctly)", "1065698": "The file exists in the input folder, I shared with you the submit notebook, and please tell me if you see something that I don't see. I tried to re-sumbit now, with an error again.\n\nhttps://www.kaggle.com/abdessalemboukil/submit\nThanks!", "1065721": "I don't see any reason yet... The only thing I would do is-if your model doesn't use lectures, can only select questions at once? (before 'preprocess(test_data, features)...'", "1065731": "It seem not to make a difference in my other notebook. This is an odd behavior." }, "source": "meta" }